Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77113.92117623411,
-      "required_pages_per_second_upper_estimate": 0.005334919995920603
+      "required_token_positions_per_second": 77332.69996667499,
+      "required_pages_per_second_upper_estimate": 0.005350055620280036
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9639.240147029264,
-      "required_pages_per_second_upper_estimate": 0.0040357485593374084
+      "required_token_positions_per_second": 9666.587495834374,
+      "required_pages_per_second_upper_estimate": 0.004047198323204526
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19278.48029405853,
-      "required_pages_per_second_upper_estimate": 0.004015199037538099
+      "required_token_positions_per_second": 19333.174991668748,
+      "required_pages_per_second_upper_estimate": 0.0040265905006541345
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28917.720441087793,
-      "required_pages_per_second_upper_estimate": 0.0017335731856407456
+      "required_token_positions_per_second": 28999.76248750312,
+      "required_pages_per_second_upper_estimate": 0.0017384914810523932
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28917.720441087793,
-      "required_pages_per_second_upper_estimate": 0.002248225568684802
+      "required_token_positions_per_second": 28999.76248750312,
+      "required_pages_per_second_upper_estimate": 0.0022546039769287688
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28917.720441087793,
-      "required_pages_per_second_upper_estimate": 0.3709254683891663
+      "required_token_positions_per_second": 28999.76248750312,
+      "required_pages_per_second_upper_estimate": 0.3719778156706959
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,14 +158,14 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.060117845320295515,
-    "producer_p95_seconds": 0.13593336100029774,
-    "gpu_update_seconds": 0.9560919594725832,
-    "producer_batches_per_second": 16.63399602351358,
-    "gpu_batches_per_second": 1.0459244951203628,
-    "producer_headroom": 15.903629851980254,
+    "producer_batch_seconds": 0.04673064564060425,
+    "producer_p95_seconds": 0.1322100339998542,
+    "gpu_update_seconds": 0.9533871186674154,
+    "producer_batches_per_second": 21.3992335498806,
+    "gpu_batches_per_second": 1.0488918723778617,
+    "producer_headroom": 20.401753615811753,
     "sustainable": true,
-    "buffer_seconds": 60.233793446772744,
+    "buffer_seconds": 60.06338847604717,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -20,3 +20,4 @@
 {"step": 927000, "loss": 0.5282454490661621, "lr": 3e-05, "grad_norm": 0.31548371961677524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135686554750, "tokens_per_second": 155026.8245825793, "data_wait_s": 7.901999651949154e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 927100, "loss": 0.521530032157898, "lr": 3e-05, "grad_norm": 0.34304400346035546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135701278418, "tokens_per_second": 154438.85215189034, "data_wait_s": 8.729999990464421e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 927200, "loss": 0.5180409550666809, "lr": 3e-05, "grad_norm": 0.3321759100936417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135715787627, "tokens_per_second": 152243.67296993933, "data_wait_s": 6.393000148818828e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 927300, "loss": 0.5016880631446838, "lr": 3e-05, "grad_norm": 0.3294710496400265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135730323364, "tokens_per_second": 152485.7517690447, "data_wait_s": 8.615999831818044e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7610177085cb1b4363d3c2c0295630b53c84cf09022b76a0ff74f03d8eecee0c
-size 46761
+oid sha256:f820d7f0f0722792ac9b8e282afb3b8417c552cc1dd3d80db22d6903a10529b0
+size 48879